Skip to content

Latest commit

 

History

History

pytorch

README.md

SWIN (Shifted Windows Vision Transformers)

Swin Transformer: Hierarchical Vision Transformer using Shifted Windows, optimised for Graphcore's IPU. Based on the models provided by SWIN

Framework Domain Model Datasets Tasks Training Inference Reference
PyTorch Vision SWIN ImageNet LSVRC 2012 Image recognition, Image classification

✅
Min. 16 IPUs (POD16) required

❌

 Swin Transformer: Hierarchical Vision Transformer using Shifted Windows

Instructions summary

  1. Install and enable the Poplar SDK (see Poplar SDK setup)

  2. Install the system and Python requirements (see Environment setup)

  3. Download the ImageNet LSVRC 2012 dataset (See Dataset setup)

Poplar SDK setup

To check if your Poplar SDK has already been enabled, run:

 echo $POPLAR_SDK_ENABLED

If no path is provided, then follow these steps:

  1. Navigate to your Poplar SDK root directory

  2. Enable the Poplar SDK with:

cd poplar-<OS version>-<SDK version>-<hash>
. enable.sh
  1. Additionally, enable PopART with:
cd popart-<OS version>-<SDK version>-<hash>
. enable.sh

More detailed instructions on setting up your Poplar environment are available in the Poplar quick start guide.

Environment setup

To prepare your environment, follow these steps:

  1. Create and activate a Python3 virtual environment:
python3 -m venv <venv name>
source <venv path>/bin/activate

  1. Navigate to the Poplar SDK root directory

  2. Install the PopTorch (PyTorch) wheel:

cd <poplar sdk root dir>
pip3 install poptorch...x86_64.whl

  1. Navigate to this example's root directory

  2. Install the Python requirements:

pip3 install -r requirements.txt
  1. Build the custom ops:
make all

More detailed instructions on setting up your PyTorch environment are available in the PyTorch quick start guide.

Dataset setup

ImageNet LSVRC 2012

Download the ImageNet LSVRC 2012 dataset from the source or via kaggle

Disk space required: 144GB

.
├── bounding_boxes
├── imagenet_2012_bounding_boxes.csv
├── train
└── validation

3 directories, 1 file

Running and benchmarking

To run a tested and optimised configuration and to reproduce the performance shown on our performance results page, use the examples_utils module (installed automatically as part of the environment setup) to run one or more benchmarks. The benchmarks are provided in the benchmarks.yml file in this example's root directory.

For example:

python3 -m examples_utils benchmark --spec <path to benchmarks.yml file>

Or to run a specific benchmark in the benchmarks.yml file provided:

python3 -m examples_utils benchmark --spec <path to benchmarks.yml file> --benchmark <name of benchmark>

For more information on using the examples-utils benchmarking module, please refer to the README.

Custom training

Precision options

We can train with fp32.32 and fp16.16 (input.weights). We can use the PRECISION parameter in config to modify the required precision

Accuracy is as follows:

model input size precision machine acc
tiny 224 16.16 pod16 80.9%
tiny 224 32.32 pod16 81.21%
tiny 224 mix v100 81.3%(SOTA)
base 224 32.32 pod16 83.5%
base 224 mix v100 83.5%(SOTA)
base 384 32.32 pod16 84.47%
base 384 mix v100 84.5%(SOTA)
large 224 32.32 pod16 86.27%
large 224 mix v100 86.3%(SOTA)

NOTE:The Acc which marked SOTA is quoted from the paper

In the above results, base 384 and large are finetune models it needs the pretrained model form imagenet21k, which require you to provide the path of the "pretrained-model" parameter to get the correct ACC

You can load the base 384 model pretrained on imagenet21k by:

wget https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_base_patch4_window7_224_22k.pth

and large 224 model pretrained on imagenet21k by:

wget https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_large_patch4_window7_224_22k.pth

Once the training finishes, you can validate accuracy:

python3 validate.py --cfg YOUR_CONFIG --checkpoint /path/to/checkpoint.pth

License

This application is licensed under Apache License 2.0. Please see the LICENSE file in this directory for full details of the license conditions.

The following files are created by Graphcore and are licensed under Apache License, Version 2.0 :

  • configs/*
  • dataset/build_ipu.py
  • models/build.py
  • options.py
  • README.md
  • swin_test.py
  • train_swin.py
  • utils.py
  • validate.py The following files are based on code from repo which is licensed under the MIT License:
  • config.py
  • dataset/cached_image_folder.py
  • dataset/samplers.py
  • dataset/zipreader.py
  • lr_scheduler.py
  • model/swin_transformer.py
  • optimizer.py
  • train_swin.sh See the headers in the files for details.

External packages:

  • transformers is licenced under Apache License, Version 2.0
  • pytest is licensed under MIT License
  • torchvision is licensed under BSD 3-Clause License